View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Elsevier - Publisher Connector

Historia Mathematics 7 (1980), 369-378

RICHARD COURANTAND THE FINITE ELEMENTMETHOD:

A FURTHERLOOK

BY FRANK WILLIAMSON, JR.

RAMAPO COLLEGE OF NEW JERSEY
MAHWAH, NJ 07430

SUMMARIES
The finite element method has in recent years
become one of the most popular and effective numer-
ical procedures for boundary value and eiqenvalue
problems. In the early 1960s it was found that the
method was an independent rediscovery of a simpler
idea proposed in 1943 by Richard Courant. The im-
portance of Courant's 1943 paper has been emphasized
by G. Stranq in 1973 and by others. This note seeks
to describe briefly Courant's finite element work
which led to his publication of 1943. It is shown
that Courant used the finite element ideas as early
as 1922 in a proof employing Dirichlet's principle.
La methode de l'element fini est devenue depuis
les dernieres an&es un des pro&d&s numeriques les
plus populaires et les plus efficaces pour les
problemes de la valeur frontier-e et pour les problemes
de valeurs propres. Au d&but des annees 60, on a
constate que cette methode Qtait une redecouverte in-
dependante d'une id&e plus simple proposee par Richard
Courant en 1943. En 1973 G. Stranq et d'autres ont
mis en relief l'importance de l'etude de Courant de
1943. L'article qui suit cherche a decrire brikve-
ment le travail de Courant sur "l'element fini" qui
aboutit a la publication de 1943. On y montre que
Courant employait dejh en 1922 l'idee de l'ele-
ment fini dans une demonstration de l'existence
selon le principe de Dirichlet.
Die Methode der finiten Elemente ist in den letzten
Jahren zu einem sehr popul;iren und effektiven numeri-
schen Verfahren zur Behandlunq von Randwert- und Eiqen-
wertaufqaben qeworden. Anfanq der sechziqer Jahre
wurde festqestellt, dass es sich dabei urn eine unab-
Mnqiqe Wiederentdeckunq einer einfacheren Idee handelt,
die Richard Courant 1943 vorqeschlaqen hatte. Auf die
Bedeutunq von Courants Arbeiten haben G. Stranq (1973)
und andere hinqewiesen. In dieser Mitteilunq wird

0315-0860/80/040369-10$02.00/O
Copyright 0 I980 by Academic Press, Inc.
All rights of reproduction in any form reserved.

369
370 Frank Williamson, Jr. HM7

Courants BeschSftiqunq mit finiten Elementen, die zur

Verijffentlichunq des Jahres 1943 fiihrte, kurz beschrie-
ben. Dabei wird nachyewiesen, dass Courant den Gedanken
finiter Elemente bereits 1922 in Verbinduny mit einem
Existenzbeweis fiir das Dirichletsche Prinzip verwendete.

1. INTRODUCTION
The finite element method was developed in the American and
European aircraft industries in the 1950s [Argyris 1960, Turner et
al., 19561 and was transformed in the 1960s from a physically based
procedure with a limited mathematical foundation into the present-
day method resting upon variational principles [Fraeijs de Veubeke
1965, 1964, Melosh 19631. The technique has become increasingly
popular with both engineers and numerical analysts, and its applica
tion has been extended far beyond the original aero-elasticity
problems [Strang and Fix 1973, Williamson 1976, Zienkiewicz 19701.
It has been recognized increasingly often as a general approxima-
tion method for boundary value and eigenvalue problems, often
superior to finite difference methods. The name "finite element
method" was first used by Clough [Clough 1960, Zienkiewicz 19701.
The present variational form of the finite element method for
boundary value problems is based upon the replacement of a govern-
ing differential equation by the equivalent extremum statement of
the problem. In this form, the solution is given by the function
which causes an appropriate integral functional, defined over the
region for the problem, to attain that extremum. An approximate
solution is obtained through the construction of a piecewise
(usually polynomial) trial function continuous on a mesh of finite
elements or subregions into which the region has been divided.
This function contains n unspecified parameters, most often re-
presenting values of the function or its derivatives at the nodal
points of the mesh. The application of the integral functional
to this trial function produces a function of the n parameters,
the extremum of which is found via n partial differentiations.
The system of n (usually linear) equations which results is solved
to fix the approximation. With proper attention to completeness of
the trial function and to other considerations, convergence with
mesh refinement can be expected as n approaches *. The procedure
for eigenvalue problems is similar.
As the variational reformulation was taking place in 196Os,
it was observed, probably as early as 1963 [Key 19741, that the
new variational form was an independent rediscovery of a simpler
method presented briefly in a [19431 paper by Richard Courant
(1888-1972). While other antecedents of the variational form
from outside the aircraft industry have been noted in the survey
literature, for example, those of Synge [19571 and Polyd 119521,
the paper by Courant [1943] is the earliest example cited [e.g.,
Kolata 1974, Strang 19731.
HM7 Courant and the Finite Element Method 371

We show that this was not the only instance of Courant's use
of the finite element procedure. Rather, it was the last in a
series of short presentations of the idea dating from 1922, when
it first appeared in an existence proof for a variation of the
Riemann Mapping Theorem [Hurwitz & Courant 19221.
To indicate the essentials of the finite element method used
in Courant's article of 1943 [f], consider the following passage
which appears after a discussion of the Rayleigh-Ritz method [21:
If the variational problems contain derivatives not
higher than the first order the method of finite
differences can be subordinated to the Rayleigh-Ritz
method by considering in the competition only func-
tions 4 which are linear in the meshes of a sub-
division of our net into triangles formed by diagonals
of the squares of the net. For such polyhedral func-
tions the integrals become sums expressed by the finite
number of values of $ in the net points and the minimum
conditions become our difference equations. Such an
interpretation suggests a wide generalization, which
provides great flexibility and seems to have consider-
able practical value [Courant 1943, 101.
It is the idea of piecewise trial functions, defined over a
mesh of subregions or finite elements, which is the central
feature of the finite element method. While later in the article
Courant applies the method to a torsion problem involving a column
with doubly connected cross section, using both regular and ir-
regular meshes of varying sizes, most of the paper is devoted to
a variety of topics other than the finite element method.

2. AN EXISTENCE PROOF
Since Courant's first mention of the finite element method
in 1922 occurred in the course of an existence proof, it is use-
ful to explain something of his approach to such proofs as il-
lustrated in [Courant 1912, 1914, 1950, Hurwitz & Courant 19221.
The goal was to show the existence of a solution to the varia-
tional problem: find the function u, selected from a prescribed
class of admissible functions defined over some domain R, which
results in a minimum value, d, for the Dirichlet integral,

Js(
i-l
(g,' + (E)*)dxdy = D(G). (A)

Courant customarily began with a sequence of functions ol, Q2f

,...,such that
$3
lim D(@j) = d; (B)
j-tm
372 Frank Williamson, Jr. HM7

this sequence was "smoothed" or "normed," creating another sequence

which was shown either to converge to the desired solution or to
provide a convergent subsequence. (That is, the original sequence wa
replaced by a sequence the terms of which are harmonic in circular
or rectangular subdomains of fi.) Since only the existence of the
original sequence IQj) was necessary, and this was usually assured
by the existence of a lower bound for D(Q), the explicit specifica-
tion of the functions in the sequence was not required. Courant
referred to a sequence with property (B) as a minimizing sequence
[eke Minimalfolge]. This general method of proof, based upon
an integral functional (A) rather than upon an equivalent partial
differential equation, is now called Dirichlet's principle. How-
ever, it was David Hilbert (1862-1943) who first made use of the
minimizing sequence in this way [Hilbert 19041. In so doing, he
rescued Dirichlet's principle, which had fallen into disuse after
K. Weierstrass, in 1870, had constructed a counterexample which
invalidated its previous applications [Monna 1975, 35-651. More-
over, Hilbert was partly responsible for Courant's interest in
the Dirichlet principle, having successfully suggested it to
Courant [Reid 1976, 291 as the subject of his dissertation
[Courant 19101.
In 1922, the well-known Funktionentheorie of Hurwitz and
Courant was published. Originally, Courant was only to have
edited the Hurwitz two-part volume on general function theory and
elliptic functions; but later he added a third part on geometric
function theory [Friedrichs 19721. In this part, Courant proved
the existence of the harmonic function u(x,y) required in his
proof of a version of the Riemann Mapping Theorem (u is the real
part of the desired mapping function f(z) = u + iv). The proof
followed, in general outline, the arguments using Dirichlet's
principle which are given above. However, because the mapping
theorem required that f(z) have a singularity at the origin re-
sulting from a term of the form l/z, u and the corresponding
admissible functions @ needed to contain a term of the type
x/(x2 + y2). For functions containing such a singularity, D($)
becomes infinite when the region of integration includes the
origin. Consequently, Courant adjusted the class of admissible
functions to allow only those of the form 0 = $ - S where S is a
given function with the prescribed singularity at the origin, and
the functions 0 are such that @ is continuous and has piecewise
continuous first derivatives. In the usual manner, Courant then
introduced an unspecified minimizing @I ,@2,@3,.... Accompany-
ing this sequence was a rather long footnote [31 :
The actual construction of such minimizing sequences,
unimportant for the bare existence proof, presents
no important difficulties. For example, if G is a
finite domain, bounded by simple curves C, we imagine
it covered by a net of triangles Ti, depending on an
HM7 Courant and the Finite Element Method 373

index j, so that the mesh grows finer with increasing

j- We now consider only functions $ or 0 = + -S where
the difference C$ - x/(x2 + y2) is a linear function in
each triangle Tj. For @ we understand the functions
corresponding to Tj constructed so that D(0) attains
its smallest value. This condition D(Q) = min. is
now a problem of the minimum of a function of a finite
number of variables, namely the integral interpreted
in its dependence on the values of 4 in the corner
points of the triangular partition; this problem is
certainly soluble and, moreover, as is easily seen,
by means of linear equations. That the correspond-
ing functions Oj actually form a minimizing sequence,
follows from the easily provable fact that every
admissible function 4 and its Dirichlet integral can
be approximated arbitrarily accurately with the help
of our construction with increasing j [Hurwitz &
Courant 1922, 3381.

This sketch for constructing a minimizing sequence certainly

reflects most of the essentials of the finite element method.
Of course, as Courant suggests, the specification of the
minimizing sequence was not needed in the existence proof it-
self, but Courant had mentioned his intention of indicating "at
least in principle, a method for the actual construction of the
functions in question" [Hurwitz & Courant 1922, 3221. Courant
was attracted to this kind of computational application of
existence proof methods after the appearance, in 1908, of a
paper by Walter Ritz which made the Rayleigh-Ritz method popular
[Ritz 1908; Reid 1976, 1143. The latter had given a numerical
approximation (without simple piecewise trial functions) of
solutions to partial differential equations by using a minimiz-
ing sequence similar to one used by Hilbert [1904] in his proof
based on Dirichlet's principle. Nevertheless, Courant did not
suggest in Funktionentheorie 119221 that the idea might be any-
thing more than an ad hoc scheme. It does not generalize easily
to integrands with derivatives of order greater than one (such
derivatives of piecewise linear functions vanish), and one of
the changes in his next exposition of the finite element approach
attempted to deal with this difficulty.

3. A NUMERICAL METHOD
In 1924, the first edition of Volume 1 of Methoden der mathe-
matischen Physik was published; in it there appeared a brief des-
cription [4] of the finite element idea [Courant & Hilbert 1924,
158-1591. This time it was presented as a general numerical
procedure (unconnected to an existence proof) for solving problems
in one independent variable in which the minimum of integrals of
the form:
374 Frank Williamson, Jr. HM7

CC)

was sought. Here Courant proposed a minimizing sequence of piece-

wise linear trial functions constructed over the interval of inte-
gration which had been divided into equal subintervals [Courant
& Hilbert 1924, 158-1591. He added that if derivatives higher
than first order occur in the integrand, the corresponding differ-
ence quotients may be employed there.
This version of the finite element method was described in
several subsequent places [Courant 1926, 1927; Courant & Hilbert
19311. The latter two reflect a change in viewpoint [7], requir-
ing that the sum

R yi+1- %)
C( F xi/ Yif ( > Ax
Ax
i=l

be minimized, rather than (C), into which the piecewise linear

trial function has been substituted. This method, closely related
to the finite element method and now called the finite difference
energy method, has found a growing number of recent applications
[Felippa 1973, l-141. It also resembles closely, as Courant him-
self pointed out, the method used by L. Euler in 1744 for the
derivation of the Euler differential equation [Euler 1744/1952,
l-801.
The only other published description of the finite element
method by Courant occurred in his paper of 1943, already dis-
cussed, in which he returned to problems containing derivatives
no higher than the first in the integrand.
4. CLOSING REMARKS
It should be emphasized that the earliest example to be
found in Courant's published works of his use of finite elements
was related to an existence proof in function theory which employed
Dirichlet's principle; i.e., it arose from a problem in pure rather
than in applied mathematics. Part of the explanation for this must
lie in Courant's appreciation of the twofold usefulness of the
minimizing sequence: as an important ingredient in existence proofs,
and also as a mechanism for practical approximations of the limit
function [Reid 1976, 1141.
Despite the current importance of the finite element method,
Courant, in the main, did not pursue this aspect of his work very
strenuously. In his 1943 paper he did not refer to his previous
work with finite elements, and many of his colleagues were unable
to recall much about Courant's interest in the subject [Friedrichs
1977, Key 1974, Lewy 1977, Neugebauer 19771. However, H. B. Keller
HM7 Courant and the Finite Element Method 375

[6] recalls that from about 1954 Courant had at least two mathe-
maticians at New York University, including Keller, working from
time to time on finite element ideas; but "unfortunately we did
not push too hard" [Keller 19791. Nevertheless, a paper based
upon piecewise linear triangular finite elements was published
by Friedrichs and Keller [1966], following a similar, earlier
paper by Friedrichs [1962].
By 1959, the difficulties with.the Rayleigh-Ritz method, as
well as with the general class of such direct methods, had been
known to applied mathematicians for some time, so that finite
difference methods were becoming more and more attractive. In
that year, Courant invited his colleagues "to give some attention
to the highly flexible Rayleigh-Ritz method and to the so-called
direct variational methods which may have become unfashionable
undeservedly" [1960]. Although it may not have appeared so at
the time, the basis for a technique like the finite element method
had been available to Courant since 1922. Later such techniques,
as developed in the 1960s and 197Os, would prove to be the answer
to his concern.

ACKNOWLEDGMENTS
The author wishes to acknowledge the help given by T. Hawkins,
I. Fried, H. B. Keller, and C. Favre-Gilly in the preparation
of this paper.

NOTES
1. Although Courant's procedure required the construction
of the linear system of equations for the nodal values, it did
not cite the efficiencies possible in the assembly of that system.
Such efficiencies account for much of the practicality of the
present variational form of the finite element method.
2. The Rayleigh-Ritz method [Ritz 19081 approximates the
solution of boundary value problems by replacing the governing
differential equation with an equivalent statement which requires
finding the extremum of an integral functional over a prescribed
class of admissible functions. This extremum is approximated by
a trial function which is a linear combination of n so-called
"coordinate" functions with the same number of unspecified para-
meters as coefficients. The trial function is inserted into the
functional which produces a function of the n parameters. The
n conditions -for the extremum of this functional yield an equal
number of equations and unknowns and thus fix the approximation.
With a proper choice of coordinate functions, the minimizing
sequence of trial functions converges to the extremum as n increases
3. This footnote did not appear in subsequent editions of
[Hurwitz & Courant 19221 with which this writer is familiar. It
may have been removed by 0. Neugebauer, who revised the first
edition [Reid 1976, 941, although he does not recall whether he
removed it or not.
376 Frank Williamson, Jr. HM7

4. The passage was almost certainly Courant's work rather

than Hilbert's [Courant & Hilbert 1924, v-vi; Reid 1976, 97-98;
Weaver 19721. It is possible that it was written before 1922
[Reid 1976, 92-931.
5. The passage written in 1931 appears in the English edition
of Methoden der mathematischen Physik [Courant & Hilbert 19531.
6. Keller also mentioned that the ideas of the penalty method,
a scheme for treating boundary conditions other than the natural
ones for a variational problem (and which Courant was the first
to employ in the 1943 article), are now applied by users of the
finite element method [Keller 19791.

REFERENCES
Arqyris, J. 1960. Energy theorems and structural analysis. London:
Butterworths.
Clouqh, R. 1960. The finite element in plane stress analysis.
In Proceedings of the Second American Society of Civil
Engineers Conference on Electronic Computation, Pittsburgh.
Courant, R. 1910. 6ber die Anwendung des Dirichletschen Prinzipes
auf die Probleme des Konformen Abbildung. Giittingen: Druck
der Univ.-Buchdruckerei von W. Fr. Kaestner.
1912. tjber die Methode des Dirichletschen Prinzipes.
Mathematische Annalen 72, 517-550.
1914. ljber die Existenztheoreme der Pontential- und
Funktionentheorie. Journal fiir die Reine und Angewandte
Mathematik 144, 190-211.
1926. iber direkte Methoden bei Variations- und Rand-
wertproblemen. Jahresbericht Deutsche Mathematiker-vereinig-
ung 34, 90-117.
1927. cber direkte Methoden in der Variationsrechnunq
und iiber Verwandte Fraqen. Mathematische Annalen 97, 711-
736.
1943. Variational methods for the solution of problems
of equilibrium and vibrations. Bulletin of the American Mathe-
matical Society 49, l-23.
1950. Dirichlet's principle, conformal mapping, and
minimal surfaces. New York: Interscience.
1960. Remarks about the Rayleigh-Ritz Method. In
Boundary problems in differential equations, R. Lanqer, ed.,
pp. 273-277. Madison: Univ. of Wisconsin Press.
Courant, R., & Hilbert D. 1924. Methoden der mathematischen
Physik, Vol. 1. Berlin: Springer.
1931. Methoden der mathematischen Physik, Vol.. 1. Berlin:
Springer.
1953. Methods of mathematical physics, Vol. 1. New York:
Interscience.
Euler, L. 1744/1952. Methodus inveniendi lineas curvas. C. Cara-
theodory, ed. A reprinting of the 1744 edition. Bern: Ore11
Fussli Turici, 1952.
HM7 Courant and the Finite Element Method 377

Fellipa, C. 1973. Finite element and finite difference energy

techniques for the numerical solution of partial difference
equations. In Proceedings, Summer Computer Simulation Con-
ference, Montreal.
Fraeijs de Veubeke, B. 1965. Displacement and equilibrium models
in the finite element method. In Stress analysis, 0. Zienkie-
wicz and G. Holister, eds., pp. 145-197. New York: Wiley.
1964. Upper and lower bounds in matrix structural
analysis. In Matrix methods of structural analysis, B. Fraeijs
de Veubeke, ed., pp. 165-197. New York: Macmillan.
Friedrichs, K. 1962. A finite difference scheme for the Neumann
and the Dirichlet problem. AEC Computing and Applied Mathe-
matics Center Report, Courant Institute of Mathematical
Sciences, New York University, NYO-9760.
1972. Richard Courant 1888-1972, Remarks delivered at
the Memorial, February 18, and at the Meeting of the Graduate
Faculty, New York University, March 15.
1977. Letter to the Author. February 3.
Friedrichs, K., & Keller, H. 1966, A finite difference scheme
for generalized Neumann problems. In Numerical solution
of partial differential equations, J. Bramble, ed., pp. l-20.
New York: Academic Press.
Hilbert, D. 1904. Faber das Dirichletsche Prinzip. Mathematische
Annalen 59, 161-186.
Hurwitz, A., & Courant, R. 1922. Funktionentheorie. Berlin:
Springer.
Keller, H. 1979. Letter to T. Hawkins. August 8.
Key, S. W. 1974. Letter to the Author. April 29.
Kolata, G. 1974. The finite element method: A mathematical
revival. Science 184, 887-889.
Lewy, H. 1977. Letter to the Author. February 1.
Melosh, R. 1963. Basis for derivation of matrices for the direct
stiffness method. Journal of the American Institute of Aeronau-
tics and Astronautics 1, 1631-1637.
Monna, A. F. 1975. Dirichlet's principle, a mathematical comedy
of errors and its influence on the development of analysis.
Utrecht: Oosthoek, Scheltema & Holkema.
Neugebauer, 0. 1977. Letter to the Author, January 31.
Poly6, G. 1952. Sur une interpretation de la m&hode des diffgr-
ences finies qui peut fournir des bornes supdrieures ou
infgrieures. Academic des Science, Paris, Comptes Rendus
235, 995-997.
Reid, C. 1976. Courant in Gijttingen and New York. New York:
Springer.
Ritz, W. 1908. iber eine neue Methode zur L&sung gewisser
Variationsprobleme der mathematischen Physik. Journal fiir
die Reine und Angewandte Mathematik 135, l-61.
Strang, G. 1973. Piecewise Polynomials and the finite element
method. Bulletin of the American Mathematical Society 79,
1128-1137.
378 Frank Williamson, Jr. HM7

Strang, G., & Fix, G. 1973. An analysis of the finite element

method. Preface. Englewood Cliffs, N.J.: Prentice-Hall.
Synge, J. 1957. The hypercircle in mathematical physics.
Cambridge: Cambridge Univ. Pres..
Turner, M., Clough, R., Martin, H., & Topp, L., 1956. Stiffness
and deflection analysis of complex structures. Journal of
the Aeronautical Sciences 23, 805-823.
Weaver, w. 1972. Richard Courant (1888-1972). Yearbook of the
American Philosophical Society, 142-149.
Williamson, C. F., Jr. 1976. A history of the finite element
method to the middle 1960s. Ph.D. dissertation, Boston
University.
Zienkiewicz, 0. 1970. The finite element method: From intuition
to generality. Applied filechanics Reviews 23, 249-256.